AI personas and synthetic data let researchers keep working with evidence after fieldwork. A model can answer another question or mimic language patterns in a segment, but its answer remains an estimate—not a late interview with the original respondent. Our first article covered the more established uses of AI-assisted analysis and in-survey probing.
Interacting with AI Personas
Have you ever finished a study and wished you could have conducted a series of qualitative interviews with a subset of survey participants? Or have budgetary, time, or feasibility limitations that have forced you to skip or curtail research?
Built from existing respondent or segment data, an AI persona can generate modeled responses in language that reflects patterns in the source material. It can be useful for workshop exploration and hypothesis-building, but it does not add another interview to the study.
Teams can use personas to:
- Explore themes in data from hard-to-reach or specialized populations.
- Support repeated questioning without adding respondent burden.
- Surface hypotheses, possible barriers, or weak ideas for validation in later research.
- Give workshop participants a more accessible way to interact with an existing evidence base.
Generating Synthetic Data
It is common to finish a study and discover a follow-up question that was never asked. Synthetic data can model a possible answer using one or more existing data sources.
This includes:
- Exploring secondary questions that were cut because of survey-length constraints.
- Estimating possible responses to new questions raised by the results.
- Testing hypotheses before investing in new fieldwork.
- Screening early ideas to identify which ones merit validation with real participants.
What the Source Data Permits
Digital personas and synthetic data can draw on multiple sources—for example, survey responses paired with sales or behavioral information, reviews, or qualitative material. More input does not automatically make the output more reliable; source quality, permissions, and validation still determine what the model can support.
Researchers should document the sources and assumptions, validate material conclusions where possible, and label modeled output when presenting it.